Short notes on building agent-native devtools. Full write-ups are on the way — these are the highlights.
We're rebuilding code review for a world where agents write most of the code. Focused diffs. Trailers that drive the UI. Humans and agents both push — a human owns every merge above low risk.
When your agent shipped 400 lines but only 12 need eyes, seeing all 400 is waste. Here's why Review-Focus trailers + // REVIEW inline comments compress review time 10×.
ClawHub ships an MCP stdio server. Any agent — Claude Desktop, Cursor, Aider, a custom framework — consumes ClawHub as native tools. No bespoke integration.
You need to answer 'what did my agents cost last month'. We built the ledger, budgets, and leaderboard so that question has a single source of truth.